x86/dsputil: port clear_block functions to yasm

Signed-off-by: James Almer <jamrial@gmail.com>
Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/x86/dsputil.asm

@@ -513,3 +513,63 @@ BSWAP32_BUF
 
 INIT_XMM ssse3
 BSWAP32_BUF
+
+;----------------------------------------
+; void ff_clear_block(int16_t *blocks);
+;----------------------------------------
+; %1 = number of xmm registers used
+; %2 = number of inline store loops
+%macro CLEAR_BLOCK 2
+cglobal clear_block, 1, 1, %1, blocks
+    ZERO  m0, m0
+%assign %%i 0
+%rep %2
+    mova  [blocksq+mmsize*(0+%%i)], m0
+    mova  [blocksq+mmsize*(1+%%i)], m0
+    mova  [blocksq+mmsize*(2+%%i)], m0
+    mova  [blocksq+mmsize*(3+%%i)], m0
+    mova  [blocksq+mmsize*(4+%%i)], m0
+    mova  [blocksq+mmsize*(5+%%i)], m0
+    mova  [blocksq+mmsize*(6+%%i)], m0
+    mova  [blocksq+mmsize*(7+%%i)], m0
+%assign %%i %%i+8
+%endrep
+    RET
+%endmacro
+
+INIT_MMX mmx
+%define ZERO pxor
+CLEAR_BLOCK 0, 2
+INIT_XMM sse
+%define ZERO xorps
+CLEAR_BLOCK 1, 1
+
+;-----------------------------------------
+; void ff_clear_blocks(int16_t *blocks);
+;-----------------------------------------
+; %1 = number of xmm registers used
+%macro CLEAR_BLOCKS 1
+cglobal clear_blocks, 1, 2, %1, blocks, len
+    add   blocksq, 768
+    mov      lenq, -768
+    ZERO       m0, m0
+.loop
+    mova  [blocksq+lenq+mmsize*0], m0
+    mova  [blocksq+lenq+mmsize*1], m0
+    mova  [blocksq+lenq+mmsize*2], m0
+    mova  [blocksq+lenq+mmsize*3], m0
+    mova  [blocksq+lenq+mmsize*4], m0
+    mova  [blocksq+lenq+mmsize*5], m0
+    mova  [blocksq+lenq+mmsize*6], m0
+    mova  [blocksq+lenq+mmsize*7], m0
+    add   lenq, mmsize*8
+    js .loop
+    RET
+%endmacro
+
+INIT_MMX mmx
+%define ZERO pxor
+CLEAR_BLOCKS 0
+INIT_XMM sse
+%define ZERO xorps
+CLEAR_BLOCKS 1

libavcodec/x86/dsputil_init.c

@@ -534,8 +534,6 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
     c->add_pixels_clamped        = ff_add_pixels_clamped_mmx;
 
     if (!high_bit_depth) {
-        c->clear_block  = ff_clear_block_mmx;
-        c->clear_blocks = ff_clear_blocks_mmx;
         c->draw_edges   = ff_draw_edges_mmx;
     }
 
@@ -547,6 +545,10 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
 #endif /* HAVE_MMX_INLINE */
 
 #if HAVE_MMX_EXTERNAL
+    if (!high_bit_depth) {
+        c->clear_block  = ff_clear_block_mmx;
+        c->clear_blocks = ff_clear_blocks_mmx;
+    }
     c->vector_clip_int32 = ff_vector_clip_int32_mmx;
 #endif /* HAVE_MMX_EXTERNAL */
 }
@@ -585,7 +587,10 @@ static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
 {
 #if HAVE_SSE_INLINE
     c->vector_clipf = ff_vector_clipf_sse;
+#endif /* HAVE_SSE_INLINE */
 
+#if HAVE_YASM
+#if HAVE_SSE_EXTERNAL
     /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
     if (CONFIG_XVMC && avctx->hwaccel && avctx->hwaccel->decode_mb)
         return;
@@ -594,9 +599,7 @@ static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
         c->clear_block  = ff_clear_block_sse;
         c->clear_blocks = ff_clear_blocks_sse;
     }
-#endif /* HAVE_SSE_INLINE */
-
-#if HAVE_YASM
+#endif
 #if HAVE_INLINE_ASM && CONFIG_VIDEODSP
     c->gmc = ff_gmc_sse;
 #endif

libavcodec/x86/dsputil_mmx.c

@@ -172,61 +172,6 @@ void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
     } while (--i);
 }
 
-#define CLEAR_BLOCKS(name, n)                           \
-void name(int16_t *blocks)                              \
-{                                                       \
-    __asm__ volatile (                                  \
-        "pxor %%mm7, %%mm7              \n\t"           \
-        "mov  $-"#n",       %%"REG_a"   \n\t"           \
-        "1:                             \n\t"           \
-        "movq %%mm7,   (%0, %%"REG_a")  \n\t"           \
-        "movq %%mm7,  8(%0, %%"REG_a")  \n\t"           \
-        "movq %%mm7, 16(%0, %%"REG_a")  \n\t"           \
-        "movq %%mm7, 24(%0, %%"REG_a")  \n\t"           \
-        "add    $32, %%"REG_a"          \n\t"           \
-        "js      1b                     \n\t"           \
-        :: "r"(((uint8_t *) blocks) + n)              \
-        : "%"REG_a);                                    \
-}
-CLEAR_BLOCKS(ff_clear_blocks_mmx, 768)
-CLEAR_BLOCKS(ff_clear_block_mmx, 128)
-
-void ff_clear_block_sse(int16_t *block)
-{
-    __asm__ volatile (
-        "xorps  %%xmm0, %%xmm0          \n"
-        "movaps %%xmm0,    (%0)         \n"
-        "movaps %%xmm0,  16(%0)         \n"
-        "movaps %%xmm0,  32(%0)         \n"
-        "movaps %%xmm0,  48(%0)         \n"
-        "movaps %%xmm0,  64(%0)         \n"
-        "movaps %%xmm0,  80(%0)         \n"
-        "movaps %%xmm0,  96(%0)         \n"
-        "movaps %%xmm0, 112(%0)         \n"
-        :: "r" (block)
-        : "memory");
-}
-
-void ff_clear_blocks_sse(int16_t *blocks)
-{
-    __asm__ volatile (
-        "xorps  %%xmm0, %%xmm0              \n"
-        "mov     $-768,         %%"REG_a"   \n"
-        "1:                                 \n"
-        "movaps %%xmm0,    (%0, %%"REG_a")  \n"
-        "movaps %%xmm0,  16(%0, %%"REG_a")  \n"
-        "movaps %%xmm0,  32(%0, %%"REG_a")  \n"
-        "movaps %%xmm0,  48(%0, %%"REG_a")  \n"
-        "movaps %%xmm0,  64(%0, %%"REG_a")  \n"
-        "movaps %%xmm0,  80(%0, %%"REG_a")  \n"
-        "movaps %%xmm0,  96(%0, %%"REG_a")  \n"
-        "movaps %%xmm0, 112(%0, %%"REG_a")  \n"
-        "add      $128,         %%"REG_a"   \n"
-        "js         1b                      \n"
-        :: "r"(((uint8_t *) blocks) + 128 * 6)
-        : "%"REG_a);
-}
-
 void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, int w)
 {
     x86_reg i = 0;